Research team:
Israel: Shaul Druckmann, Albert Gidon, Prof. Idan Segev and collaborators
Switzerland: Felix Schuermann, Henry Markram and collaborators
Background
The brain consists of approximately 1012 neurons. If each person alive today would have a thousand children, the number of people on earth would be equal to the number of neurons within a single brain. The Neocortex is an area of the brain that is most evolved in higher mammals. There is growing experimental evidence that the neocortex is organized in terms of its architecture into column like structures arranged in grid like formation, each consisting of ten to a hundred thousand cells. It has been suggested that each such column has a specific computational function, for instance processing the information from a specific whisker in a rat's mustache. Thus, it is tempting to think of the cortical column as a basic universal computational unit of the cortex.
The field of computational neuroscience deals amongst other things with the processing of information in neural networks. In the past decades there has been a growing understanding of the processing carried out by the single neuron.
Model of a nerve cell
Indeed, very detailed understanding of the electrical activity within a single neuron has been achieved by mathematical modeling of this activity. Yet so far, the activity of neuronal networks has been described in more abstract fashion, partly at least due to the computational resources required to solve the great number of equations describing the detailed activity of such networks.
Research Questions:
Do we know enough to reproduce the complex electrical activity of a large scale cortical network? If not what do we need to know better?
How do the different components of cortical networks, such as cell types, topology and receptors interact to create the network behavior?
Can we relate the observed differences between normal and pathological brain activity to specific causes?
Research Methods:
We use an IBM BlueGene/L supercomputer with over 8000 parallel processors. What takes the computer 1 day to simulate would take 8000 days or 20 years to simulate on a standard desktop. All in all, the network contains on the order of ten thousand neurons connected by ten billion synapses. Each processor simulates from one neuron to a few. Hence, we have the computational power to simulate each neuron and synapse in realistic detail.
We developed the tools to be able to generate the network model each time from scratch by directly drawing the necessary data from experimental databases. Thus, we can update the databases and generate a more refined model as often as new data becomes available. In addition, we can highlight the areas in which experimental data is lacking.
The Model:
We model data collected over more than a decade in the labs of prof. Henry Markram and collaborators. It mainly consists of electrophysiological recordings of single cell activity in rat somatosensory cortex. We are able to generate a model of one or a few cortical columns, simulate its activity in different scenarios and visualize the result. We are able to reproduce experimentally observed activity in several levels of resolution (single cell, layer, etc') and are continually seeking to find more experimental calibration methods and apply them to refine our model.
image of a simulated network in action
Applications:
There are an increasing number of studies concerning the difference in brain activity in healthy and pathological brains. Occasionally, physical differences between healthy networks and pathological ones can be observed that might cause the difference in behavior. However, it is extremely difficult experimentally to separate the main cause for pathology from side effects and compensatory mechanisms. Yet, in the model one can check the differences one by one and attempt to elucidate which is the main cause. In addition, one can examine the affectivity of potential treatments through simulating their effect on the model network.
Discussion and Conclusions:
The increasing power of computers affords both an intriguing opportunity and a great challenge for theoretical neuroscience. Using new generations of super computers neural networks can be simulated in unprecedented detail. The analysis of these simulations, unraveling cause and effect in the normal and pathological activity of neural networks will be a major challenge that we have begun to address.
We show that detailed models can be generated directly from available experimental data bases. The resulting model can be simulated, visualized and interrogated in detail. Its calibration by comparison to available experiments is an ongoing effort in Israel and in Switzerland.
More about... Research Tools - The equipment and tools at a researcher's disposal are important in determining the research questions that can be addressed. Some questions can only be explored in laboratories fitted with advanced equipment; a simpler facility would be unable to handle the job. In brain research, for example, imaging devices allow researchers to "peek" into the brain to examine its functioning-something that perceptual experiments do not. Using advanced computerized equipment for theoretical research, scientists can construct computer images of complex, dynamic systems that cannot be studied using simple computers or paper-and-pencil techniques. Particularly complex endeavors, such as the Blue Brain project, require both enormous funding and the joint cooperation of several laboratories. Nevertheless, advanced equipment is no guarantee of success. Researchers must plan their methods wisely to make the most efficient use of the tools available to them. Limited resources, in some cases, can actually encourage creativity and lead to scientific breakthroughs.
